AI Glossary
Plain-language definitions for the terms used across the Prompt Hub.
Prompt
A one-off instruction you copy and paste into an AI tool to get a response.
System Prompt
A background instruction that shapes how an AI behaves in a session, often invisible to the end user.
Skill
A reusable instruction file for one repeatable job — trigger it, get a consistent result.
SKILL.md
A Markdown file containing a skill's complete instructions, quality rules, guardrails, and test commands.
Agent
A named AI skill with a clear role — e.g. "CV Strategist" or "Post Writer".
AI Team
Multiple AI agents or skills working together on a shared goal.
Workflow
The ordered steps that produce a complete outcome using one or more skills.
Workflow Pack
A bundled set of skills, context files, and quality gates for a specific use case.
Context File
A document that gives your AI memory, brand rules, career details, or project context.
Memory
Persistent information an AI carries between sessions — often provided via a context file.
Quality Gate
A skill that checks output against standards before you publish, send, deploy, or trust it.
Guardrail
An instruction that tells the AI what to avoid — e.g. "do not invent statistics".
Tool Calling
A capability that lets an AI call external tools, APIs, or services during a task.
MCP
Model Context Protocol — a standard that lets AI tools connect to external apps and services.
Automation
A workflow connected to tools or APIs so it runs with minimal human input.
Human Approval
A checkpoint where a human must review and approve AI output before it is acted on.
Hallucination
When an AI generates information that sounds plausible but is invented or incorrect.
RAG
Retrieval-Augmented Generation — giving an AI access to external documents to reduce hallucination.
Long-context Model
An AI model that can handle very large inputs — useful for documents, codebases, or research.
Multimodal Model
An AI that can process multiple input types — text, images, audio, or video.
Local LLM
An AI language model that runs on your own machine, not in the cloud.
Trigger Phrase
The exact phrase you use to activate a skill or agent — e.g. "Run the LinkedIn Post Writer".
Pass Criteria
The specific standards an AI output must meet to be considered acceptable.
Prompt Linter
A tool that scans prompts and skills for quality issues, missing fields, and safety gaps.
Skill Health Score
A 0–100 score reflecting how complete, safe, and test-ready a skill is.